Car Price Prediction

ai_generated_photo_turbo_az.png

Table of contents

  • 1.Importing Libaries

  • 2.Importing Data

  • 3.Cleaning Data

    • 3.1.Fixing Inconsistencies
    • 3.2.Dealing with Missing Values
    • 3.3.Removing Duplicates

  • 4.Exploratory Data Analysis (EDA)

    • 4.1.Univariate Analysis
    • 4.2.Bivariate Analysis
    • 4.3.Geospatial Analysis🌍



  • 5.Feature Engineering
    • 5.1.Creating new columns and Dropping unecessary ones
    • 5.2.Outliers
    • 5.3.Dimensionality reduction
    • 5.4.Encoding



  • 6.Model Building

Importing Libraries

¶

In [1]:
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
import geopandas as gpd
import folium
from sklearn.model_selection import train_test_split
from xgboost import XGBRegressor
from datetime import datetime
from sklearn.metrics import mean_squared_error, r2_score
import osmnx as ox

Importing data

¶

In [3]:
df = pd.read_excel('turbo_az_dataset.xlsx',names = ['City', 'Price', 'Make', 'Model', 'Year', 'Body_Type', 'Color', 'Engine', 'Kilometrage', 'Transmission', 'Drivetrain', 'Is_New', 'Number_of_Seats', 'Condition', 'Market', 'Damaged', 'Owners'])
df.head()
Out[3]:
City Price Make Model Year Body_Type Color Engine Kilometrage Transmission Drivetrain Is_New Number_of_Seats Condition Market Damaged Owners
0 Bərdə 8 500 AZN GAZ 53 1992 Yük maşını Yaş Asfalt 4.3 L / 150 a.g. / Benzin 500 000 km Mexaniki (MT) Arxa Xeyr NaN Vuruğu yoxdur, rənglənməyib NaN NaN NaN
1 Tovuz 8 500 AZN Mercedes C 180 1994 Sedan Gümüşü 1.8 L / 122 a.g. / Benzin 596 890 km Avtomat (AT) Arxa Xeyr 5 Vuruğu yoxdur, rənglənib Amerika NaN NaN
2 Ağdaş 2 500 AZN LADA (VAZ) 2108 1989 Hetçbek, 3 qapı Sarı 1.5 L / 78 a.g. / Benzin 172 820 km Mexaniki (MT) Ön Xeyr 5 Vuruğu var, rənglənib NaN NaN NaN
3 Sumqayıt 28 900 AZN BMW 528 2012 Sedan Ağ 2.0 L / 245 a.g. / Benzin 152 410 km Avtomat (AT) Arxa Xeyr 5 Vuruğu yoxdur, rənglənməyib Amerika NaN NaN
4 Bakı 21 500 AZN Hyundai ix35 2012 Offroader / SUV, 5 qapı Ağ 2.0 L / 150 a.g. / Benzin 270 000 km Avtomat (AT) Ön Xeyr 5 Vuruğu yoxdur, rənglənməyib Rəsmi diler NaN NaN
In [4]:
df.shape
Out[4]:
(20078, 17)
In [5]:
df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 20078 entries, 0 to 20077
Data columns (total 17 columns):
 #   Column           Non-Null Count  Dtype 
---  ------           --------------  ----- 
 0   City             20078 non-null  object
 1   Price            20078 non-null  object
 2   Make             20078 non-null  object
 3   Model            20078 non-null  object
 4   Year             20078 non-null  int64 
 5   Body_Type        20078 non-null  object
 6   Color            20078 non-null  object
 7   Engine           20078 non-null  object
 8   Kilometrage      20078 non-null  object
 9   Transmission     20078 non-null  object
 10  Drivetrain       20078 non-null  object
 11  Is_New           20078 non-null  object
 12  Number_of_Seats  17893 non-null  object
 13  Condition        20077 non-null  object
 14  Market           16334 non-null  object
 15  Damaged          103 non-null    object
 16  Owners           8014 non-null   object
dtypes: int64(1), object(16)
memory usage: 2.6+ MB

Data Cleaning

¶

3.1 Fixing Inconsistencies¶

In [8]:
df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 20078 entries, 0 to 20077
Data columns (total 17 columns):
 #   Column           Non-Null Count  Dtype 
---  ------           --------------  ----- 
 0   City             20078 non-null  object
 1   Price            20078 non-null  object
 2   Make             20078 non-null  object
 3   Model            20078 non-null  object
 4   Year             20078 non-null  int64 
 5   Body_Type        20078 non-null  object
 6   Color            20078 non-null  object
 7   Engine           20078 non-null  object
 8   Kilometrage      20078 non-null  object
 9   Transmission     20078 non-null  object
 10  Drivetrain       20078 non-null  object
 11  Is_New           20078 non-null  object
 12  Number_of_Seats  17893 non-null  object
 13  Condition        20077 non-null  object
 14  Market           16334 non-null  object
 15  Damaged          103 non-null    object
 16  Owners           8014 non-null   object
dtypes: int64(1), object(16)
memory usage: 2.6+ MB

Let's look for inconsistencies in dataset. We can clearly see that when some columns should be numeric but they are in object data type. Such as Price , Engine, Kilometrage, Number_of_Seats. Let's transform them to numeric columns first.

Price column

In price column some values are in different currencies. We need to bring them to a single currency and change the column to a numeric dtype.

In [12]:
df['Price']
Out[12]:
0                     8 500 AZN
1                     8 500 AZN
2                     2 500 AZN
3                    28 900 AZN
4                    21 500 AZN
                  ...          
20073                14 300 AZN
20074                19 800 AZN
20075                34 000 AZN
20076    ≈ 86 360 AZN50 800 USD
20077                18 500 AZN
Name: Price, Length: 20078, dtype: object
In [13]:
# keep prices only in AZN currency
df['Price'] = df['Price'].str.split('AZN').str[0]
In [14]:
# removing any extra spaces and characters just to keep numbers
df['Price'] = df['Price'].str.replace(r'[^0-9]','', regex = True)
df['Price']= df['Price'].astype(int)
df['Price']
Out[14]:
0         8500
1         8500
2         2500
3        28900
4        21500
         ...  
20073    14300
20074    19800
20075    34000
20076    86360
20077    18500
Name: Price, Length: 20078, dtype: int32

Engine column

In [16]:
df['Engine']
Out[16]:
0        4.3 L / 150 a.g. / Benzin
1        1.8 L / 122 a.g. / Benzin
2         1.5 L / 78 a.g. / Benzin
3        2.0 L / 245 a.g. / Benzin
4        2.0 L / 150 a.g. / Benzin
                   ...            
20073    1.8 L / 204 a.g. / Benzin
20074    1.5 L / 184 a.g. / Benzin
20075    1.6 L / 183 a.g. / Benzin
20076    1.6 L / 230 a.g. / Hibrid
20077    1.5 L / 160 a.g. / Benzin
Name: Engine, Length: 20078, dtype: object

We see that Engine column combines data for 3 different values (engine, engine power,fuel type). We should separate it into 3 distinct columns.

In [18]:
df[['Engine(L)','Engine_Power','Fuel_Type']] = df['Engine'].str.split('/',expand = True)
In [19]:
df.loc[df['Make'] == 'Tesla',['Engine(L)','Engine_Power','Fuel_Type']]
Out[19]:
Engine(L) Engine_Power Fuel_Type
331 283 a.g. Elektro None
341 770 a.g. Elektro None
406 345 a.g. Elektro None
699 1020 a.g. Elektro None
711 450 a.g. Elektro None
759 554 a.g. Elektro None
817 462 a.g. Elektro None
844 450 a.g. Elektro None
1893 283 a.g. Elektro None
2111 562 a.g. Elektro None
2134 450 a.g. Elektro None
2478 562 a.g. Elektro None
2861 258 a.g. Elektro None
2920 462 a.g. Elektro None
3051 351 a.g. Elektro None
3065 428 a.g. Elektro None
3168 351 a.g. Elektro None
3229 321 a.g. Elektro None
4094 534 a.g. Elektro None
4122 462 a.g. Elektro None
5014 450 a.g. Elektro None
5775 351 a.g. Elektro None
6198 510 a.g. Elektro None
6756 351 a.g. Elektro None
7250 500 a.g. Elektro None
7885 462 a.g. Elektro None
8631 845 a.g. Elektro None
9190 351 a.g. Elektro None
9834 450 a.g. Elektro None
9905 450 a.g. Elektro None
10055 258 a.g. Elektro None
10262 450 a.g. Elektro None
10335 462 a.g. Elektro None
10556 450 a.g. Elektro None
10697 283 a.g. Elektro None
11587 450 a.g. Elektro None
11597 562 a.g. Elektro None
11707 670 a.g. Elektro None
12636 462 a.g. Elektro None
12965 560 a.g. Elektro None
14044 260 a.g. Elektro None
14684 690 a.g. Elektro None
14745 283 a.g. Elektro None
14759 845 a.g. Elektro None
15708 462 a.g. Elektro None
15718 351 a.g. Elektro None
15822 562 a.g. Elektro None
15891 462 a.g. Elektro None
16550 470 a.g. Elektro None
16601 462 a.g. Elektro None
16979 450 a.g. Elektro None
17321 462 a.g. Elektro None
17460 462 a.g. Elektro None
17580 283 a.g. Elektro None
17846 283 a.g. Elektro None
17890 304 a.g. Elektro None
19411 1020 a.g. Elektro None
19494 540 a.g. Elektro None
19511 450 a.g. Elektro None
19643 462 a.g. Elektro None

For the electrical cars values of the three columns above are mixed. This problem needs to be fixed.

In [21]:
df.loc[df['Engine'].str.contains('Elektro'),'Fuel_Type'] = 'Electric'
df.loc[df['Engine'].str.contains('Elektro'),'Engine_Power'] = df['Engine(L)']
df.loc[df['Engine'].str.contains('Elektro'),'Engine(L)'] = '0 L'

Engine Power column

In [23]:
df['Engine(L)'] = df['Engine(L)'].str.replace(r'[a-zA-Z\s]','',regex = True).astype(float)
In [24]:
df.loc[df['Engine'].str.contains('sm3'),'Engine(L)'] =df.loc[df['Engine'].str.contains('sm3'),'Engine(L)'] /1000

Converted sm3 to liters

Kilometrage, Engine_Power(a.g) and Number_of_Seats columns

In [27]:
df
Out[27]:
City Price Make Model Year Body_Type Color Engine Kilometrage Transmission Drivetrain Is_New Number_of_Seats Condition Market Damaged Owners Engine(L) Engine_Power Fuel_Type
0 Bərdə 8500 GAZ 53 1992 Yük maşını Yaş Asfalt 4.3 L / 150 a.g. / Benzin 500 000 km Mexaniki (MT) Arxa Xeyr NaN Vuruğu yoxdur, rənglənməyib NaN NaN NaN 4.3 150 a.g. Benzin
1 Tovuz 8500 Mercedes C 180 1994 Sedan Gümüşü 1.8 L / 122 a.g. / Benzin 596 890 km Avtomat (AT) Arxa Xeyr 5 Vuruğu yoxdur, rənglənib Amerika NaN NaN 1.8 122 a.g. Benzin
2 Ağdaş 2500 LADA (VAZ) 2108 1989 Hetçbek, 3 qapı Sarı 1.5 L / 78 a.g. / Benzin 172 820 km Mexaniki (MT) Ön Xeyr 5 Vuruğu var, rənglənib NaN NaN NaN 1.5 78 a.g. Benzin
3 Sumqayıt 28900 BMW 528 2012 Sedan Ağ 2.0 L / 245 a.g. / Benzin 152 410 km Avtomat (AT) Arxa Xeyr 5 Vuruğu yoxdur, rənglənməyib Amerika NaN NaN 2.0 245 a.g. Benzin
4 Bakı 21500 Hyundai ix35 2012 Offroader / SUV, 5 qapı Ağ 2.0 L / 150 a.g. / Benzin 270 000 km Avtomat (AT) Ön Xeyr 5 Vuruğu yoxdur, rənglənməyib Rəsmi diler NaN NaN 2.0 150 a.g. Benzin
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
20073 Bakı 14300 Mercedes C 250 2008 Sedan Ağ 1.8 L / 204 a.g. / Benzin 248 850 km Avtomat (AT) Arxa Xeyr 5 Vuruğu yoxdur, rənglənməyib Amerika NaN NaN 1.8 204 a.g. Benzin
20074 Bakı 19800 Ford Fusion (North America) 2017 Offroader / SUV, 5 qapı Yaş Asfalt 1.5 L / 184 a.g. / Benzin 142 000 km Avtomat (AT) Ön Xeyr 5 Vuruğu yoxdur, rənglənməyib Digər NaN NaN 1.5 184 a.g. Benzin
20075 Bakı 34000 Hyundai Sonata 2020 Sedan Qara 1.6 L / 183 a.g. / Benzin 70 000 km Avtomat (AT) Ön Xeyr 5 Vuruğu yoxdur, rənglənib Amerika NaN NaN 1.6 183 a.g. Benzin
20076 Bakı 86360 Hyundai Grandeur 2023 Sedan Qara 1.6 L / 230 a.g. / Hibrid 25 000 km Avtomat (AT) Ön Xeyr NaN Vuruğu yoxdur, rənglənməyib NaN NaN NaN 1.6 230 a.g. Hibrid
20077 Bakı 18500 Chevrolet Malibu 2017 Sedan Ağ 1.5 L / 160 a.g. / Benzin 179 264 km Avtomat (AT) Ön Xeyr 5 Vuruğu yoxdur, rənglənməyib Amerika NaN NaN 1.5 160 a.g. Benzin

20078 rows × 20 columns

In [28]:
df[['Kilometrage','Engine_Power']] = df[['Kilometrage','Engine_Power']].apply(
    lambda x: x.str.replace(r'[^0-9]','',regex = True).astype(int)
)
In [29]:
df['Number_of_Seats'] = pd.to_numeric(df['Number_of_Seats'], errors = 'coerce')
In [30]:
df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 20078 entries, 0 to 20077
Data columns (total 20 columns):
 #   Column           Non-Null Count  Dtype  
---  ------           --------------  -----  
 0   City             20078 non-null  object 
 1   Price            20078 non-null  int32  
 2   Make             20078 non-null  object 
 3   Model            20078 non-null  object 
 4   Year             20078 non-null  int64  
 5   Body_Type        20078 non-null  object 
 6   Color            20078 non-null  object 
 7   Engine           20078 non-null  object 
 8   Kilometrage      20078 non-null  int32  
 9   Transmission     20078 non-null  object 
 10  Drivetrain       20078 non-null  object 
 11  Is_New           20078 non-null  object 
 12  Number_of_Seats  17700 non-null  float64
 13  Condition        20077 non-null  object 
 14  Market           16334 non-null  object 
 15  Damaged          103 non-null    object 
 16  Owners           8014 non-null   object 
 17  Engine(L)        20078 non-null  float64
 18  Engine_Power     20078 non-null  int32  
 19  Fuel_Type        20078 non-null  object 
dtypes: float64(2), int32(3), int64(1), object(14)
memory usage: 2.8+ MB

Now we see that Price , Engine(L), Kilometrage, Engine_Power(a.g), Number_of_Seats columns are in numeric data type.

Dealing with missing values¶

In [33]:
df.isnull().sum()
Out[33]:
City                   0
Price                  0
Make                   0
Model                  0
Year                   0
Body_Type              0
Color                  0
Engine                 0
Kilometrage            0
Transmission           0
Drivetrain             0
Is_New                 0
Number_of_Seats     2378
Condition              1
Market              3744
Damaged            19975
Owners             12064
Engine(L)              0
Engine_Power           0
Fuel_Type              0
dtype: int64
In [34]:
null_cols = df.isnull().mean()
null_cols.loc[null_cols>0]
Out[34]:
Number_of_Seats    0.118438
Condition          0.000050
Market             0.186473
Damaged            0.994870
Owners             0.600857
dtype: float64

There are null values in 5 columns. Let's go one by one.

Number_of_seats column

In [37]:
df['Number_of_Seats'].isnull().sum()
Out[37]:
2378
In [38]:
df_train = df.dropna(subset = ['Number_of_Seats']).copy()
df_predict = df.loc[df['Number_of_Seats'].isnull()].copy()
In [39]:
X_train = df_train[['Make', 'Model', 'Body_Type']]
y_train = df_train['Number_of_Seats']
X_train_encoded = pd.get_dummies(X_train, drop_first=True)
In [41]:
X_predict = df_predict[['Make', 'Model', 'Body_Type']]
X_predict_encoded = pd.get_dummies(X_predict, drop_first=True)
X_predict_encoded = X_predict_encoded.reindex(columns = X_train_encoded.columns,fill_value = 0) #ensuring consistency with thr trainig dataset
In [42]:
model = RandomForestClassifier(n_estimators=100, random_state=42)
model.fit(X_train_encoded,y_train)
Out[42]:
RandomForestClassifier(random_state=42)
In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
RandomForestClassifier(random_state=42)
In [43]:
predicted_seats = model.predict(X_predict_encoded)
df.loc[df['Number_of_Seats'].isnull(),'Number_of_Seats'] = predicted_seats

Market column

In [45]:
df['Market'].isnull().sum()
Out[45]:
3744
In [46]:
df['Market'].unique()
Out[46]:
array([nan, 'Amerika', 'Rəsmi diler', 'Yaponiya', 'Avropa', 'Rusiya',
       'Koreya', 'Çin', 'Digər', 'Dubay'], dtype=object)
In [47]:
df_market_train = df.dropna(subset = ['Market']).copy()
df_market_predict = df.loc[df['Market'].isnull()].copy()
In [48]:
X_train = df_market_train[['Make', 'Model']]
y_train = df_market_train['Market']
X_train_encoded = pd.get_dummies(X_train, drop_first = True)
#y_train_encoded = pd.get_dummies(y_train, drop_first = True)
In [49]:
X_predict = df_market_predict[['Make', 'Model']]
X_predict_encoded = pd.get_dummies(X_predict , drop_first = True)
X_predict_encoded = X_predict_encoded.reindex(columns = X_train_encoded.columns,fill_value = 0)
In [50]:
model = RandomForestClassifier(n_estimators=100,random_state = 42)
model.fit(X_train_encoded,y_train)
Out[50]:
RandomForestClassifier(random_state=42)
In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
RandomForestClassifier(random_state=42)
In [51]:
predicted_markets = model.predict(X_predict_encoded)
In [52]:
df.loc[df['Market'].isnull(),'Market'] = predicted_markets

Condition column

In [54]:
df['Condition'].isnull().sum()
Out[54]:
1
In [55]:
df.loc[df['Condition'].isnull()]
Out[55]:
City Price Make Model Year Body_Type Color Engine Kilometrage Transmission Drivetrain Is_New Number_of_Seats Condition Market Damaged Owners Engine(L) Engine_Power Fuel_Type
18224 Bakı 19500 Ssang Yong Korando 2013 Offroader / SUV, 5 qapı Ağ 2.0 L / 149 a.g. / Benzin 196000 Avtomat (AT) Tam Xeyr 5.0 NaN Rəsmi diler NaN 2 2.0 149 Benzin

There is only 1 missing row. So , it is better to drop that row.

In [57]:
df = df.dropna(subset = 'Condition')

Damaged and Owners columns

These 2 columns will be dropped because more than 50% is null

In [60]:
df['Damaged'].value_counts()
Out[60]:
Damaged
Bəli    103
Name: count, dtype: int64
In [61]:
damaged_cars = df.loc[df['Damaged'] == 'Bəli']
In [62]:
plt.figure(figsize=(10, 6))
damaged_cars['Make'].value_counts().plot(kind = 'bar')
Out[62]:
<Axes: xlabel='Make'>
No description has been provided for this image

From the visual, we can see that most of the damaged cars for sale are Mercedes, Opel, Lada (VAZ), Hyundai and Ford.

In [64]:
df = df.drop(columns = ['Damaged','Owners'])

Removing Duplicates¶

In [66]:
df.duplicated().sum()
Out[66]:
162
In [67]:
# dropping duplicates
df.drop_duplicates(inplace = True)

Exploratory data analysis

Univariate Analysis¶

In [70]:
df.describe()
Out[70]:
Price Year Kilometrage Number_of_Seats Engine(L) Engine_Power
count 19915.000000 19915.000000 1.991500e+04 19915.000000 19915.000000 19915.000000
mean 32006.302234 2012.264524 1.856518e+05 5.030178 2.226406 185.939292
std 42656.455048 9.199316 1.549953e+05 0.905965 1.426442 106.016526
min 500.000000 1906.000000 0.000000e+00 1.000000 0.000000 3.000000
25% 11850.000000 2007.000000 7.500000e+04 5.000000 1.500000 114.000000
50% 21500.000000 2013.000000 1.700000e+05 5.000000 2.000000 164.000000
75% 35500.000000 2019.000000 2.600000e+05 5.000000 2.400000 224.000000
max 777479.000000 2025.000000 1.635680e+06 7.000000 20.003000 1548.000000

Insights:
Price: The average price for a car in Azerbaijan is around 32 006 AZN. However the median price is 21 500 AZN, which is much lower than mean. It means data is skewed heavily to the right. Price range is very large, from 500 AZN to 777 479 AZN.

Year: The oldest car for sale on turbo.az is from 1906 and the newest is from 2025. The average car in this dataset is from the year 2012.

Engine(L): Average engine size is around 2.23 liters. There are also electrical cars listed with 0 L engines and car(s) with very large engine size up to 20 L.

Kilometrage: The mean kilometrage for cars are 185K km and highest one is 1.635 mln km. There also new cars as well with 0 kilometrage.

Engine_Power: The average horsepower is around 185. Most powerful car has a power of 1548 while the least power car has a power 3.

In [74]:
df[['Price','Kilometrage','Year','Engine_Power','Engine(L)']].hist(bins=10, figsize=(10, 6))
plt.show()
No description has been provided for this image

From the the distributions we see that many columns have positive skewness.
Price - we can say majority of the cars are in lower price and there are a few and very expensive cars.
Year - Year of the majority of the cars are new (from 2000) with a smaller number of very old cars.
Kilometrage, Engine_Power, Engine(L) - the same as price column, thye have right skewness. There are very few cars with high kilometrage, engine size and power.

In [76]:
df['Make'].value_counts().head(10).plot(kind = 'bar',figsize = (18,6))
plt.xlabel('Make')
plt.title('Top 10 Most Common Car Makes')
plt.show()
No description has been provided for this image

Top 3 car brands in sale on turbo.az are Mercedes, Hyundai and Toyata

In [78]:
regions = df['City'].value_counts()
more_cars_regions = regions[regions > 150]
more_cars_regions.plot(kind = 'pie', figsize = (6,6),fontsize = 12.5, 
                                     explode = [0.05]*len(more_cars_regions), 
                                     ylabel = '', 
                                     legend=True, 
                                     startangle= 160, autopct='%1.1f%%')
plt.title('Cars by Region with More Than 150',fontsize=15, color = '#006400')
plt.show()
No description has been provided for this image

Regions with most of the cars are Bakı, Sumqayıt, Gəncə,Lənkəran and Xırdalan

In [80]:
df.columns
Out[80]:
Index(['City', 'Price', 'Make', 'Model', 'Year', 'Body_Type', 'Color',
       'Engine', 'Kilometrage', 'Transmission', 'Drivetrain', 'Is_New',
       'Number_of_Seats', 'Condition', 'Market', 'Engine(L)', 'Engine_Power',
       'Fuel_Type'],
      dtype='object')
In [81]:
plt.subplot(1,2,1)
df['Body_Type'].value_counts().plot(kind = 'bar', figsize = (18,6),fontsize = 12.5,color = '#737373')
plt.title('Cars by Body Type',fontsize=15)

plt.subplot(1,2,2)
df['Color'].value_counts().plot(kind = 'bar', figsize = (18,6),fontsize = 12.5,color = '#08659f')
plt.title('Cars by Color', fontsize=15)
plt.show()
No description has been provided for this image

The following inferences can be made from the above bar plots:

Body Type of most of the cars are 'Sedan', 'Offroader / SUV', 'Hetchbek', 'Liftbek' and 'Universal'
And color of most of the cars are 'Ağ', 'Qara', 'Gümüşü', 'Yaş Asfalt'and 'Göy'

In [83]:
plt.subplot(1,2,1)
df['Fuel_Type'].value_counts().plot(kind = 'bar', figsize = (18,6),fontsize = 12.5,color = '#737373')
plt.title('Cars by Fuel Type',fontsize=15)

plt.subplot(1,2,2)
df['Transmission'].value_counts().plot(kind = 'bar', figsize = (18,6),fontsize = 12.5,color = '#08659f')
plt.title('Cars by Transmission', fontsize=15)
plt.show()
No description has been provided for this image

The following inferences can be made from the above bar plots:

Body Type of most of the cars are 'Sedan', 'Offroader / SUV', 'Hetchbek', 'Liftbek' and 'Universal'
And color of most of the cars are 'Ağ', 'Qara', 'Gümüşü', 'Yaş Asfalt'and 'Göy'

Bivariate Analysis¶

In [85]:
plt.figure(figsize = (15,6))
sns.boxplot(data = df, x = 'Fuel_Type', y = 'Price')
plt.title('Price Distribution by Fuel Type',fontsize=15)
plt.show()
No description has been provided for this image
  • Median price for Electric and dizel-hibrid cars are higher than others. So, it means these cars are usually more expensive than others.
  • The box for Benzin and Qaz categories are smaller than others. It means for these categories 50% of data concentrated among small price range (difference between min and max price) is higher.
  • There are outliers across almost all types. It mean in each category there are very expensive cars for all fuel types.
  • Benzin fuel type has the highest range.

Geospatial Analysis¶

In [263]:
az_regions_gdf = gpd.read_file('azrbaycan_rayonlar.geojson')
az_regions_gdf.head(2)
Out[263]:
OBJECTID Name_AZ Name PARENT_ID Code_OSM Name_RU Area Shape_Leng Shape_Le_1 ID Shape_Le_2 Shape_Length Shape_Area Upper_case geometry
0 1 Qax Gakh 1 1204 Гах 1487.563667 3.145374 3.145374 403.0 3.145374 3.145374 0.159953 QAX POLYGON ((47.01821 41.5551, 47.01816 41.55467,...
1 2 Zaqatala Zagatala 1 1204 Закатала 1354.885693 2.853307 2.853307 402.0 2.853307 2.853307 0.146303 ZAQATALA MULTIPOLYGON (((46.47881 41.64208, 46.47853 41...

Data source: https://opendata.az/

In [265]:
az_regions_gdf = az_regions_gdf.loc[~az_regions_gdf.Name_AZ.isin(['Nərimanov', 'Nizami', 'Pirallahı','Qaradağ', 'Sabunçu', 
                                                'Səbail', 'Suraxanı', 'Xətai', 'Xəzər',
                             'Yasamal', 'Binəqədi', 'Nəsimi'])]
In [245]:
baku_geom = ox.geocode_to_gdf('Baku, Azerbaijan')
water_gdf = ox.features_from_place("Baku, Azerbaijan", tags={'natural':'water'}) # water polygon

water_gdf_proj = water_gdf.to_crs(epsg=32639) # change crs for calculating area
water_gdf_proj['area'] = water_gdf_proj['geometry'].area

caspian_gdf = water_gdf_proj[water_gdf_proj['area'] > 1e10]  #  Filter large water bodies (Caspian Sea)
caspian_gdf = caspian_gdf.to_crs(baku_geom.crs) 

baku_land = baku_geom.copy() 
baku_land['geometry'] = baku_land['geometry'].difference(
    gpd.GeoSeries(caspian_gdf.geometry).union_all()
) # Remove Caspian Sea from Baku polygon
In [267]:
baku_geom = baku_land['geometry']
sumgait_geom = ox.geocode_to_gdf('Sumgait, Azerbaijan')['geometry']
In [269]:
baku = pd.DataFrame({'Name_AZ': 'Bakı', 'geometry': baku_geom})
sumgait = pd.DataFrame({'Name_AZ': 'Sumqayıt', 'geometry': sumgait_geom})
In [271]:
az_regions_gdf = pd.concat([az_regions_gdf[['Name_AZ','geometry']],baku,khirdalan,sumgait])
In [303]:
df_agg = df.groupby('City').agg(
    count_car = ('Price','count'),
    avg_price = ('Price','mean')
).reset_index()
df_agg['Total_Car_Count'] = df_agg['count_car'].apply(lambda x: f"{x:,}") 
df_agg['Average_Price'] =  df_agg['avg_price'].apply(lambda x: f"{x:,.0f} AZN")
In [307]:
df_agg.loc[df_agg.City == 'Xırdalan','City'] = 'Abşeron'
In [309]:
choropleth_gdf  = az_regions_gdf.merge(df_agg, how = 'left',left_on = 'Name_AZ', right_on = 'City')
In [317]:
choropleth_gdf.isnull().sum()
Out[317]:
Name_AZ             0
geometry            0
City               15
count_car          15
avg_price          15
Total_Car_Count     0
Average_Price       0
dtype: int64
In [315]:
choropleth_gdf[['Total_Car_Count','Average_Price']] = choropleth_gdf[['Total_Car_Count','Average_Price']].fillna(0)
In [319]:
m = folium.Map(location=[40.1, 47.5], zoom_start=7)

# creating the interactive choropleth map
folium.Choropleth(
    geo_data=choropleth_gdf,
    data=choropleth_gdf,
    columns=['Name_AZ', 'count_car'],
    key_on='feature.properties.Name_AZ',
    fill_color='plasma',
    fill_opacity=0.7,
    line_opacity=0.2,
    legend_name='Average Car Price (AZN)'
).add_to(m)

# adding tooltips for interaction
style_function = lambda x: {'fillColor': '#ffffff', 'color':'#000000', 'fillOpacity': 0.1, 'weight': 0.1}
highlight_function = lambda x: {'fillColor': '#000000', 'color':'#000000', 'fillOpacity': 0.50, 'weight': 0.1}
NIL = folium.features.GeoJson(
    choropleth_gdf,
    style_function=style_function,
    control=False,
    highlight_function=highlight_function,
    tooltip=folium.features.GeoJsonTooltip(
        fields=['Name_AZ', 'Average_Price', 'Total_Car_Count'],
        aliases=['City:', 'Average Price:', 'Total Number of Cars:'],
        style=("background-color: white; color: #333333; font-family: arial; font-size: 12px; padding: 10px;")
    )
)
m.add_child(NIL)
m.save("car_price_map.html")
m
Out[319]:
Make this Notebook Trusted to load map: File -> Trust Notebook

Feature Engineering

Creating new columns and Dropping unecessary ones¶

In [96]:
df['Age_Car'] = datetime.now().year - df['Year']
df['Price_Per_Kilometer'] = df['Kilometrage']/df['Price']
In [97]:
df = df.drop(columns = ['Year','Engine','Color'])
In [98]:
sns.heatmap(df.select_dtypes(include ='number').corr(), annot = True, cmap = 'RdBu')
Out[98]:
<Axes: >
No description has been provided for this image

The following inferences can be made from correlation heatmap:¶

Price is highly (positively) correlated to the Engine_power(0.63)


Year is very negatively correlated to `Price_Per_kilometer` (-0.62)
Kilometrage is highly correlated to `Price_Per_kilometer` and Age_car(both 0.6)
Since XGboost model is generally not affected by correlations, we are not going to delete all highly correlated columns.

Outliers¶

In [101]:
normalized_price = np.log1p(df['Price'])

XGBoost is not sensitive to outliers but log transformation is good for handling with skewed data and it helps a model learn more stable relationships.

In [103]:
# Let's compare Price distribution before and after

fig, ax = plt.subplots(1,2, figsize = (15,6))
sns.histplot(df['Price'],ax = ax[0], kde = True)
ax[0].set_title('Original Price Distribution')

sns.histplot(normalized_price, ax = ax[1], kde = True)
ax[1].set_title('Normalized Price Distribution')
Out[103]:
Text(0.5, 1.0, 'Normalized Price Distribution')
No description has been provided for this image
In [104]:
df['Price'] = normalized_price
In [105]:
df['Kilometrage'] = np.log1p(df['Kilometrage'])
In [106]:
df.columns
Out[106]:
Index(['City', 'Price', 'Make', 'Model', 'Body_Type', 'Kilometrage',
       'Transmission', 'Drivetrain', 'Is_New', 'Number_of_Seats', 'Condition',
       'Market', 'Engine(L)', 'Engine_Power', 'Fuel_Type', 'Age_Car',
       'Price_Per_Kilometer'],
      dtype='object')

Dimensionality reduction¶

In [108]:
df.City.value_counts()
Out[108]:
City
Bakı        14934
Sumqayıt     1344
Gəncə         683
Lənkəran      203
Xırdalan      165
            ...  
Xızı            2
Göytəpə         2
Laçın           1
Horadiz         1
Babək           1
Name: count, Length: 65, dtype: int64
In [109]:
city_counts = df['City'].value_counts()
low_frequency_cities = city_counts[city_counts<100]
df['City'] = df['City'].apply(lambda x: 'Other_City' if x in low_frequency_cities else x)
In [110]:
df.head()
# this is final df before encoding
Out[110]:
City Price Make Model Body_Type Kilometrage Transmission Drivetrain Is_New Number_of_Seats Condition Market Engine(L) Engine_Power Fuel_Type Age_Car Price_Per_Kilometer
0 Bərdə 9.047939 GAZ 53 Yük maşını 13.122365 Mexaniki (MT) Arxa Xeyr 2.0 Vuruğu yoxdur, rənglənməyib Rusiya 4.3 150 Benzin 33 58.823529
1 Other_City 9.047939 Mercedes C 180 Sedan 13.299490 Avtomat (AT) Arxa Xeyr 5.0 Vuruğu yoxdur, rənglənib Amerika 1.8 122 Benzin 31 70.222353
2 Other_City 7.824446 LADA (VAZ) 2108 Hetçbek, 3 qapı 12.060012 Mexaniki (MT) Ön Xeyr 5.0 Vuruğu var, rənglənib Rusiya 1.5 78 Benzin 36 69.128000
3 Sumqayıt 10.271631 BMW 528 Sedan 11.934336 Avtomat (AT) Arxa Xeyr 5.0 Vuruğu yoxdur, rənglənməyib Amerika 2.0 245 Benzin 13 5.273702
4 Bakı 9.975855 Hyundai ix35 Offroader / SUV, 5 qapı 12.506181 Avtomat (AT) Ön Xeyr 5.0 Vuruğu yoxdur, rənglənməyib Rəsmi diler 2.0 150 Benzin 13 12.558140

Encoding¶

In [112]:
df = pd.get_dummies(data = df, columns = ['City', 'Make', 'Model', 'Body_Type', 'Transmission', 'Drivetrain', 
                                      'Is_New', 'Condition', 'Market', 'Fuel_Type'], drop_first = True)

Model Building

In [114]:
X = df.drop(columns = 'Price')
y = df['Price']
In [115]:
X_train,X_test, y_train, y_test = train_test_split(X,y,test_size = 0.2,random_state = 99)
In [116]:
# we will use XGBoost Regressor model
model = XGBRegressor(n_estimators=1000, learning_rate=0.05, max_depth=5, 
                     subsample=0.7, colsample_bytree=0.7, random_state=42)
In [117]:
model.fit(X_train, y_train)
Out[117]:
XGBRegressor(base_score=None, booster=None, callbacks=None,
             colsample_bylevel=None, colsample_bynode=None,
             colsample_bytree=0.7, device=None, early_stopping_rounds=None,
             enable_categorical=False, eval_metric=None, feature_types=None,
             feature_weights=None, gamma=None, grow_policy=None,
             importance_type=None, interaction_constraints=None,
             learning_rate=0.05, max_bin=None, max_cat_threshold=None,
             max_cat_to_onehot=None, max_delta_step=None, max_depth=5,
             max_leaves=None, min_child_weight=None, missing=nan,
             monotone_constraints=None, multi_strategy=None, n_estimators=1000,
             n_jobs=None, num_parallel_tree=None, ...)
In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
XGBRegressor(base_score=None, booster=None, callbacks=None,
             colsample_bylevel=None, colsample_bynode=None,
             colsample_bytree=0.7, device=None, early_stopping_rounds=None,
             enable_categorical=False, eval_metric=None, feature_types=None,
             feature_weights=None, gamma=None, grow_policy=None,
             importance_type=None, interaction_constraints=None,
             learning_rate=0.05, max_bin=None, max_cat_threshold=None,
             max_cat_to_onehot=None, max_delta_step=None, max_depth=5,
             max_leaves=None, min_child_weight=None, missing=nan,
             monotone_constraints=None, multi_strategy=None, n_estimators=1000,
             n_jobs=None, num_parallel_tree=None, ...)
In [118]:
predictions = model.predict(X_test)
In [119]:
mse = mean_squared_error(y_test, predictions)
r2 = r2_score(y_test, predictions)

print(f"Mean Squared Error (MSE): {mse:.2f}")
print(f"R-squared (R2): {r2:.2f}")

Mean Squared Error (MSE): 0.01
R-squared (R2): 0.99

Model explains 99% of the variation in car prices.
And MSE is 0.01 which is very low value shows that the model's price predictions are on average extremely close to the actual prices.